Tim Williams tube compressor

[rankot]

Anybody tried to build this?



I don't have tube diode so I can't breadboard it, tried to simulate in LTspice, but it doesn't seem to compress.

[Phoenix]

What pentode model are you using in spice? I haven't seen too many true pentode models, rather tetrode models - which in conventional applications behave similarly enough to true pentode, but will not work at all in this design. Even the tetrode models do not model screen behaviour even close to accurately.

The 6HZ6 pentode specified is a special case, with dual control grids (it's not really a "pentode" as such), a normal pentode model will not work at all.

All that said, I'm not sure how well this circuit would really work.

[rankot]

I've used EF80_g3, which is a model with all five electodes, so it is modelled like a real pentode.

In the meantime, I made it "working" in LTS with ECH83 (heptode) model, but I am not sure if it is OK. I have few of those tubes at hand so I will try to breadboard it in few days/weeks, depending on my spare time. I hope it will work, since it could be a nice single-tube compressor!

[PRR]

> I don't have tube diode

Use ordinary Si/Ge diodes. I am sure if Tim had them, he would have used them.

[thermionix]

Is it safe to put 150VDC on a breadboard?

[amptramp]

The 6HZ6 is a dual-control pentode used as an FM quadrature detector in television sets.  Grid 1 has a transconductance of 3400 µmhos and grid 3 has a transconductance of 600 µmhos with 100 volts @ 3.3 mA on grid 2 and a plate supply of 150 volts with a plate current of 3.3 mA.  The cathode bias resistor is 180 ohms to get this, not the 330 shown in the OP's schematic.  Grid 1 cuts off at -4.5 volts and grid 3 cuts off at -7 volts where cutoff is defined at 20 µA plate current.

I'm not a fan of breadboards at the best of times.  It always seems like component leads are too thin to stay in or too thick to go in at all and having things fall apart at 150 volts is not nice.  Add in tube socket to breadboard adaptors and the whole thing gets fragile.  It looks like it will work so just build it.

[PRR]

A good dry hardwood breadboard is fine to many hundred Volts.

The rating on the few solderless breadboard which have ratings is "36V". This is obviously very conservative, but how much?

[Rob Strand]

What pentode model are you using in spice? I haven't seen too many true pentode models, rather tetrode models - which in conventional applications behave similarly enough to true pentode, but will not work at all in this design. Even the tetrode models do not model screen behaviour even close to accurately.

Most tube models can't be trusted.

Back in 2001 I spent a lot of time with the "excem library" which came with a PDF describing the equations.  The models had to be translated/rewritten for different Spice versions.   I believe it had Triode, Pentode and Tetrode models.   For the Pentode there was a lot of boundary cases not handled by the basic equations.   It took a long time fixing them to the point where the simulation looked correct.   I think I tweaked some of the models parameters to match the tube datasheets.  I never got around to rolling the fixes out for the Tetrodes.

[Rob Strand]

I don't have tube diode so I can't breadboard it, tried to simulate in LTspice, but it doesn't seem to compress.

Simplify the circuit.
Start with just the pentode part.
Feed an AC signal into the audio input.
Feed a fix DC voltage or a ramp into the control input and see what happens on the AC output.
Play around with range of the DC control voltage (min/max) until you can see some form of gain control.

[rankot]

Lots of cool advices, thanks! What do you think about using ECH83 instead of 6HZ6? It seems possible to me, and I have few of them. I can't find 6HZ6 anywhere around, but maybe EF86 could be at the same spot?

[PRR]

> Most tube models can't be trusted.

They can; within limits (not usually stated).

I'm abusing a sim 12AU7. It plots fine in the range of hundreds-Volts and 10mA down to 1mA. Below a part-mA the sim stops looking "like a tube" and more like a math function. Basically I am off the published plot, nobody actually experimented on a real 12AU7, it can't be trusted beyond the blurry lines in the old manual.

A few tube models do NOT model positive G1 current, a real gotcha. Many do: as an ideal diode and a 1K-2K resistor. This DOES quickly tell you when your sim has gone positive-grid, which is usually a no-no in small audio. (Assuming you don't drive G1 with SPICE's infinite voltage source.) The real value of grid conduction is much more complicated, meaning any but the lightest class 2 sim is probably wrong.

The models I have do not do G2 well.

I had not seen G3 modeled, and wonder if the model really does something with Vg3 or it is just a place-holder for a wiring diagram.

The action of G3 varies a *lot* across different tubes. A few were bent-up so G3 has significant effect, in others it has very little effect.

I think this needs to be breadboarded (with 300V stuff, perhaps not 0.1" solderless) and several tubes tried with real signal.

[Rob Strand]

I'm abusing a sim 12AU7. It plots fine in the range of hundreds-Volts and 10mA down to 1mA.

I was probably being a bit harsh.  The triodes are the least problematic. Especially if you don't reach a boundary case.

A few tube models do NOT model positive G1 current, a real gotcha. Many do: as an ideal diode and a 1K-2K resistor. This DOES quickly tell you when your sim has gone positive-grid, which is usually a no-no in small audio. (Assuming you don't drive G1 with SPICE's infinite voltage source.) The real value of grid conduction is much more complicated, meaning any but the lightest class 2 sim is probably wrong.

Yes, it get's harder.  I used a number of "soft" diode models and the best one was based on a simple power law. The CRO waveforms looked believable.  IIRC there's not a lot of data for this region.  Also my measurements for a number of tubes showed a fair bit of variation.  My memory is a bit fuzzy on the details but at the end of the day the soft diode was a pretty good match and it only required a couple of easily tune-able parameters.   I might have chucked out series resistors altogether.

The behaviour of the plate current while the grid becomes positive is a little more dubious.    For the plate current, it was OK to stay on the triode power-law for the most part, even for a positive grid.  When Vp becomes very low (or even negative), or if Vp is less than the positive Vg I suspect the true behaviour becomes very complex.    Trying to handle these cases means more parameters, or completely changing the form of the equations.  Not to mention more parameters to fit based on limited data.   MOSFETs have similar problems, there's zillions of models with very complicated ideas behind them (which attempt to model the secondary behaviours).

The models I have do not do G2 well.

The pentodes is where it's starts to fall apart.  There are cases where the models mathematically fall apart and the "correct" way to fix them so they match the tube behaviour isn't clear.  Most of my time was spent pondering and fixing these cases.  I got things to the point where biasing with different plate and screen voltages gave good numbers.  I could simulate a saturated push-pull output stage.  It worked when Vp drops below Vscreen and when Vp drops below 100V or so where you get that "soft" clipping behaviour.    IIRC the screen current looked OK too.  No weird behaviours.

I had not seen G3 modeled, and wonder if the model really does something with Vg3 or it is just a place-holder for a wiring diagram.
The action of G3 varies a *lot* across different tubes. A few were bent-up so G3 has significant effect, in others it has very little effect.

At the time I did look at that stuff.  It's very complex behaviour.    From what I could see there wasn't a lot of info on modelling it.   I pondered over it for about a day or so and realized it needed a very detail understanding of the physics.   If the "tube-masters" of the day couldn't come-up with much then there's not much hope for me.  Besides that there's only limit info in the data sheets and limited access to samples (the job wasn't to do tube modelling measurements).  It's a problem for a university thesis, but for 50 years ago!

When I played around with this stuff my goal was only to simulate an AB push-push stage to the point where the main behaviours were captured.   The aim definitely wasn't to come up with tube models.  I tried to use as much of existing work already out there that I could.   Even so, I ended up spent a lot of time finding cracks and patching them up.

I think this needs to be breadboarded (with 300V stuff, perhaps not 0.1" solderless) and several tubes tried with real signal.

Very true.  You can't beat the real thing.    The models aren't trustworthy enough.  You might see the basic idea but the details will be off.

[printer2]

For what it is worth I have used breadboards up to 350 V, the placement of parts need to be done intelligently. If one strip will be the grid of the tube and the next the cathode (no cathode follower though) resistor the spacing should be fine. But when going north of 50-100V then leave a strip in between the conducting strips. If you have 3-400 V then two strips. I did have a metal plate with pots and switches mounted with solid wire of the proper size to use in the holes. I had another plate with tube sockets on the other side of the board. This all mounted on a piece of plywood with an area for transformers and a strip of screw terminals to build up the power supply filtering.

I need to dive into the spice pond some day but I have a full complement of resistor values and caps, it doesn't take too long to build up a real circuit to see what really gives. Now if I get around to cleaning off m. bench so that I have more than six inches to work on I might be inclined to do some more electronic work. Seems like the only time I have is for doing wood projects now.

[amptramp]

There is some pretty good 6HZ6 data here:

https://frank.pocnet.net/sheets/127/6/6HZ6.pdf

[PRR]

Thanks. This thread stirred old memory. Valley and Wallman published curves and offered theory for thermionic bottles at low current.

Below some point (below the "usual zone"), a 6SN7 works much like a 2N2222, if the '2222 worked at 1050 degrees. 1050/300 is 3.5, Shockley's Law for transistors is like 60mV/decade, 60mV*(1050/300) or 60mV*3.5 is 210mV.

I need to re-plot this for Vp and Vg and compare to my sim.

Oh: 6Y6 is oddball because its grid loses control. True of all tubes but 6Y6 must be an outlier.

[Rob Strand]

Valley and Wallman published curves and offered theory for thermionic bottles at low current.

Thanks for that.   Really cool info in that whole section.  Stuff that's not in the datasheets.
If I ever go back to tweaking tube models I'll see if I can incorporate some of those ideas.